Mixing and memory: emergent cooperation in an information marketplace

Abstract

An information marketplace consists of a group of agents buying and selling information content and services. Self-interested agents may find it rational to cheat others in the market. The authors have modeled federations of digital libraries selling information to one another. In the model, cooperation emerges through endogenous incentives. They had hypothesized that a number of factors would encourage cooperation: greater familiarity with trading partners, more memory for modeling others, fewer initially misbehaving agents, lower rates of strategy exploration, smaller world sizes, and more rapidly evolving measures of the likelihood of other agents to cooperate. The simulations confirmed many of the hypotheses. However the benefits of a small world size are dependent on memory size, and the usefulness of rapidly evolving estimates of cooperation is dependent on the types of strategies employed by the agents. They suggest that it will become increasingly important to allow agents conceptually to carve their networked environment into subcommunities where familiarity and trust can emerge, as opposed to promoting indiscriminate interactions within a large society of digital agents.